Resistance to Contextual Strategies of Power

Power appears in many faces, but it discreetly appears in the systems and infrastructures we use daily: public roadways, computer software, among others. Contextual strategies align users, systems and corresponding activity according to designers' prescribed outcomes and desired goals (Pfaffenberger, 1992). Contextual strategies shape how a design constituency allocates power. Robert Moses designed New York’s highway system to limit public buses from traveling through tunnels based on size restrictions. The limited dimensionality separated white, middle-class patrons with automobiles from public transit riders, including persons of color. Moses’ strategy of choice was exclusion based on race and class, similar to Jim Crow-segregated bathrooms, restaurants, schools and neighborhoods. The potency of Moses’ highways, however, existed not in legal code, but in the implicit values embedded in its architecture (Winner, 1980; Lessig, 2006). Design constituents may enact multiple contextual strategies to achieve prescribed outcomes and desired goals. Generative artificial intelligence models are trained on billions of text and image data across the Internet, capable of producing human-like responses and visual imaginaries at the prompt of the user. Generative models enact multiple contextual strategies. Delegation presumes a shortcoming in the expected user, and uses the technology to improve the shortcoming and increase desired outcomes. For example, seatbelts were implemented to increase safety outcomes for drivers and passengers and reduce vehicular fatalities (Latour, 1999). Centralization creates singular control over system and outputs. Standardization creates conformity and regimented outcomes. Deflection distracts impact constituents from underlying problems with a system’s implementation. In the context of generative models, we observe these strategies in action. The delegation of creative expression from a manual task model to an algorithmic task model presupposes the user’s lack of time or ability; writing and drawing become “optimized” through user prompts and algorithmic inferences. Developers centralize control over creative works through cloud servers and data policies, which routinely assume all user contributions as training data. Creative outputs become standardized and homogenized through aesthetic politics and reinforcement learning. Finally, users are encouraged to “look away” from negative externalities─ data pollution, encroaching data centers, community health risks─ and towards supposed benefits. These strategies provoke particular forms of resistance: data poisoning, forum platforms for human artists, and software counterartifacts. This paper explicates how designers use contextual strategies to allocate power, and how resisters subvert these strategies through alternative institutions and practices (Lilja, 2022; Sorenson, 2016). By identifying these strategies, we may better understand resistance, refusal and non-use practices.